Spaces:
Build error
Build error
File size: 23,463 Bytes
28c256d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 |
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
from typing import List, Optional, Sequence, Tuple
import torch
import torch.nn as nn
from mmcv.cnn import ConvModule, Scale
from mmengine.structures import InstanceData
from torch import Tensor
from mmdet.registry import MODELS
from mmdet.utils import (ConfigType, InstanceList, MultiConfig, OptConfigType,
OptInstanceList, reduce_mean)
from ..task_modules.prior_generators import anchor_inside_flags
from ..utils import images_to_levels, multi_apply, unmap
from .anchor_head import AnchorHead
@MODELS.register_module()
class ATSSHead(AnchorHead):
"""Detection Head of `ATSS <https://arxiv.org/abs/1912.02424>`_.
ATSS head structure is similar with FCOS, however ATSS use anchor boxes
and assign label by Adaptive Training Sample Selection instead max-iou.
Args:
num_classes (int): Number of categories excluding the background
category.
in_channels (int): Number of channels in the input feature map.
pred_kernel_size (int): Kernel size of ``nn.Conv2d``
stacked_convs (int): Number of stacking convs of the head.
conv_cfg (:obj:`ConfigDict` or dict, optional): Config dict for
convolution layer. Defaults to None.
norm_cfg (:obj:`ConfigDict` or dict): Config dict for normalization
layer. Defaults to ``dict(type='GN', num_groups=32,
requires_grad=True)``.
reg_decoded_bbox (bool): If true, the regression loss would be
applied directly on decoded bounding boxes, converting both
the predicted boxes and regression targets to absolute
coordinates format. Defaults to False. It should be `True` when
using `IoULoss`, `GIoULoss`, or `DIoULoss` in the bbox head.
loss_centerness (:obj:`ConfigDict` or dict): Config of centerness loss.
Defaults to ``dict(type='CrossEntropyLoss', use_sigmoid=True,
loss_weight=1.0)``.
init_cfg (:obj:`ConfigDict` or dict or list[dict] or
list[:obj:`ConfigDict`]): Initialization config dict.
"""
def __init__(self,
num_classes: int,
in_channels: int,
pred_kernel_size: int = 3,
stacked_convs: int = 4,
conv_cfg: OptConfigType = None,
norm_cfg: ConfigType = dict(
type='GN', num_groups=32, requires_grad=True),
reg_decoded_bbox: bool = True,
loss_centerness: ConfigType = dict(
type='CrossEntropyLoss',
use_sigmoid=True,
loss_weight=1.0),
init_cfg: MultiConfig = dict(
type='Normal',
layer='Conv2d',
std=0.01,
override=dict(
type='Normal',
name='atss_cls',
std=0.01,
bias_prob=0.01)),
**kwargs) -> None:
self.pred_kernel_size = pred_kernel_size
self.stacked_convs = stacked_convs
self.conv_cfg = conv_cfg
self.norm_cfg = norm_cfg
super().__init__(
num_classes=num_classes,
in_channels=in_channels,
reg_decoded_bbox=reg_decoded_bbox,
init_cfg=init_cfg,
**kwargs)
self.sampling = False
self.loss_centerness = MODELS.build(loss_centerness)
def _init_layers(self) -> None:
"""Initialize layers of the head."""
self.relu = nn.ReLU(inplace=True)
self.cls_convs = nn.ModuleList()
self.reg_convs = nn.ModuleList()
for i in range(self.stacked_convs):
chn = self.in_channels if i == 0 else self.feat_channels
self.cls_convs.append(
ConvModule(
chn,
self.feat_channels,
3,
stride=1,
padding=1,
conv_cfg=self.conv_cfg,
norm_cfg=self.norm_cfg))
self.reg_convs.append(
ConvModule(
chn,
self.feat_channels,
3,
stride=1,
padding=1,
conv_cfg=self.conv_cfg,
norm_cfg=self.norm_cfg))
pred_pad_size = self.pred_kernel_size // 2
self.atss_cls = nn.Conv2d(
self.feat_channels,
self.num_anchors * self.cls_out_channels,
self.pred_kernel_size,
padding=pred_pad_size)
self.atss_reg = nn.Conv2d(
self.feat_channels,
self.num_base_priors * 4,
self.pred_kernel_size,
padding=pred_pad_size)
self.atss_centerness = nn.Conv2d(
self.feat_channels,
self.num_base_priors * 1,
self.pred_kernel_size,
padding=pred_pad_size)
self.scales = nn.ModuleList(
[Scale(1.0) for _ in self.prior_generator.strides])
def forward(self, x: Tuple[Tensor]) -> Tuple[List[Tensor]]:
"""Forward features from the upstream network.
Args:
x (tuple[Tensor]): Features from the upstream network, each is
a 4D-tensor.
Returns:
tuple: Usually a tuple of classification scores and bbox prediction
cls_scores (list[Tensor]): Classification scores for all scale
levels, each is a 4D-tensor, the channels number is
num_anchors * num_classes.
bbox_preds (list[Tensor]): Box energies / deltas for all scale
levels, each is a 4D-tensor, the channels number is
num_anchors * 4.
"""
return multi_apply(self.forward_single, x, self.scales)
def forward_single(self, x: Tensor, scale: Scale) -> Sequence[Tensor]:
"""Forward feature of a single scale level.
Args:
x (Tensor): Features of a single scale level.
scale (:obj: `mmcv.cnn.Scale`): Learnable scale module to resize
the bbox prediction.
Returns:
tuple:
cls_score (Tensor): Cls scores for a single scale level
the channels number is num_anchors * num_classes.
bbox_pred (Tensor): Box energies / deltas for a single scale
level, the channels number is num_anchors * 4.
centerness (Tensor): Centerness for a single scale level, the
channel number is (N, num_anchors * 1, H, W).
"""
cls_feat = x
reg_feat = x
for cls_conv in self.cls_convs:
cls_feat = cls_conv(cls_feat)
for reg_conv in self.reg_convs:
reg_feat = reg_conv(reg_feat)
cls_score = self.atss_cls(cls_feat)
# we just follow atss, not apply exp in bbox_pred
bbox_pred = scale(self.atss_reg(reg_feat)).float()
centerness = self.atss_centerness(reg_feat)
return cls_score, bbox_pred, centerness
def loss_by_feat_single(self, anchors: Tensor, cls_score: Tensor,
bbox_pred: Tensor, centerness: Tensor,
labels: Tensor, label_weights: Tensor,
bbox_targets: Tensor, avg_factor: float) -> dict:
"""Calculate the loss of a single scale level based on the features
extracted by the detection head.
Args:
cls_score (Tensor): Box scores for each scale level
Has shape (N, num_anchors * num_classes, H, W).
bbox_pred (Tensor): Box energies / deltas for each scale
level with shape (N, num_anchors * 4, H, W).
anchors (Tensor): Box reference for each scale level with shape
(N, num_total_anchors, 4).
labels (Tensor): Labels of each anchors with shape
(N, num_total_anchors).
label_weights (Tensor): Label weights of each anchor with shape
(N, num_total_anchors)
bbox_targets (Tensor): BBox regression targets of each anchor with
shape (N, num_total_anchors, 4).
avg_factor (float): Average factor that is used to average
the loss. When using sampling method, avg_factor is usually
the sum of positive and negative priors. When using
`PseudoSampler`, `avg_factor` is usually equal to the number
of positive priors.
Returns:
dict[str, Tensor]: A dictionary of loss components.
"""
anchors = anchors.reshape(-1, 4)
cls_score = cls_score.permute(0, 2, 3, 1).reshape(
-1, self.cls_out_channels).contiguous()
bbox_pred = bbox_pred.permute(0, 2, 3, 1).reshape(-1, 4)
centerness = centerness.permute(0, 2, 3, 1).reshape(-1)
bbox_targets = bbox_targets.reshape(-1, 4)
labels = labels.reshape(-1)
label_weights = label_weights.reshape(-1)
# classification loss
loss_cls = self.loss_cls(
cls_score, labels, label_weights, avg_factor=avg_factor)
# FG cat_id: [0, num_classes -1], BG cat_id: num_classes
bg_class_ind = self.num_classes
pos_inds = ((labels >= 0)
& (labels < bg_class_ind)).nonzero().squeeze(1)
if len(pos_inds) > 0:
pos_bbox_targets = bbox_targets[pos_inds]
pos_bbox_pred = bbox_pred[pos_inds]
pos_anchors = anchors[pos_inds]
pos_centerness = centerness[pos_inds]
centerness_targets = self.centerness_target(
pos_anchors, pos_bbox_targets)
pos_decode_bbox_pred = self.bbox_coder.decode(
pos_anchors, pos_bbox_pred)
# regression loss
loss_bbox = self.loss_bbox(
pos_decode_bbox_pred,
pos_bbox_targets,
weight=centerness_targets,
avg_factor=1.0)
# centerness loss
loss_centerness = self.loss_centerness(
pos_centerness, centerness_targets, avg_factor=avg_factor)
else:
loss_bbox = bbox_pred.sum() * 0
loss_centerness = centerness.sum() * 0
centerness_targets = bbox_targets.new_tensor(0.)
return loss_cls, loss_bbox, loss_centerness, centerness_targets.sum()
def loss_by_feat(
self,
cls_scores: List[Tensor],
bbox_preds: List[Tensor],
centernesses: List[Tensor],
batch_gt_instances: InstanceList,
batch_img_metas: List[dict],
batch_gt_instances_ignore: OptInstanceList = None) -> dict:
"""Calculate the loss based on the features extracted by the detection
head.
Args:
cls_scores (list[Tensor]): Box scores for each scale level
Has shape (N, num_anchors * num_classes, H, W)
bbox_preds (list[Tensor]): Box energies / deltas for each scale
level with shape (N, num_anchors * 4, H, W)
centernesses (list[Tensor]): Centerness for each scale
level with shape (N, num_anchors * 1, H, W)
batch_gt_instances (list[:obj:`InstanceData`]): Batch of
gt_instance. It usually includes ``bboxes`` and ``labels``
attributes.
batch_img_metas (list[dict]): Meta information of each image, e.g.,
image size, scaling factor, etc.
batch_gt_instances_ignore (list[:obj:`InstanceData`], Optional):
Batch of gt_instances_ignore. It includes ``bboxes`` attribute
data that is ignored during training and testing.
Defaults to None.
Returns:
dict[str, Tensor]: A dictionary of loss components.
"""
featmap_sizes = [featmap.size()[-2:] for featmap in bbox_preds]
assert len(featmap_sizes) == self.prior_generator.num_levels
device = cls_scores[0].device
anchor_list, valid_flag_list = self.get_anchors(
featmap_sizes, batch_img_metas, device=device)
cls_reg_targets = self.get_targets(
anchor_list,
valid_flag_list,
batch_gt_instances,
batch_img_metas,
batch_gt_instances_ignore=batch_gt_instances_ignore)
(anchor_list, labels_list, label_weights_list, bbox_targets_list,
bbox_weights_list, avg_factor) = cls_reg_targets
avg_factor = reduce_mean(
torch.tensor(avg_factor, dtype=torch.float, device=device)).item()
losses_cls, losses_bbox, loss_centerness, \
bbox_avg_factor = multi_apply(
self.loss_by_feat_single,
anchor_list,
cls_scores,
bbox_preds,
centernesses,
labels_list,
label_weights_list,
bbox_targets_list,
avg_factor=avg_factor)
bbox_avg_factor = sum(bbox_avg_factor)
bbox_avg_factor = reduce_mean(bbox_avg_factor).clamp_(min=1).item()
losses_bbox = list(map(lambda x: x / bbox_avg_factor, losses_bbox))
return dict(
loss_cls=losses_cls,
loss_bbox=losses_bbox,
loss_centerness=loss_centerness)
def centerness_target(self, anchors: Tensor, gts: Tensor) -> Tensor:
"""Calculate the centerness between anchors and gts.
Only calculate pos centerness targets, otherwise there may be nan.
Args:
anchors (Tensor): Anchors with shape (N, 4), "xyxy" format.
gts (Tensor): Ground truth bboxes with shape (N, 4), "xyxy" format.
Returns:
Tensor: Centerness between anchors and gts.
"""
anchors_cx = (anchors[:, 2] + anchors[:, 0]) / 2
anchors_cy = (anchors[:, 3] + anchors[:, 1]) / 2
l_ = anchors_cx - gts[:, 0]
t_ = anchors_cy - gts[:, 1]
r_ = gts[:, 2] - anchors_cx
b_ = gts[:, 3] - anchors_cy
left_right = torch.stack([l_, r_], dim=1)
top_bottom = torch.stack([t_, b_], dim=1)
centerness = torch.sqrt(
(left_right.min(dim=-1)[0] / left_right.max(dim=-1)[0]) *
(top_bottom.min(dim=-1)[0] / top_bottom.max(dim=-1)[0]))
assert not torch.isnan(centerness).any()
return centerness
def get_targets(self,
anchor_list: List[List[Tensor]],
valid_flag_list: List[List[Tensor]],
batch_gt_instances: InstanceList,
batch_img_metas: List[dict],
batch_gt_instances_ignore: OptInstanceList = None,
unmap_outputs: bool = True) -> tuple:
"""Get targets for ATSS head.
This method is almost the same as `AnchorHead.get_targets()`. Besides
returning the targets as the parent method does, it also returns the
anchors as the first element of the returned tuple.
"""
num_imgs = len(batch_img_metas)
assert len(anchor_list) == len(valid_flag_list) == num_imgs
# anchor number of multi levels
num_level_anchors = [anchors.size(0) for anchors in anchor_list[0]]
num_level_anchors_list = [num_level_anchors] * num_imgs
# concat all level anchors and flags to a single tensor
for i in range(num_imgs):
assert len(anchor_list[i]) == len(valid_flag_list[i])
anchor_list[i] = torch.cat(anchor_list[i])
valid_flag_list[i] = torch.cat(valid_flag_list[i])
# compute targets for each image
if batch_gt_instances_ignore is None:
batch_gt_instances_ignore = [None] * num_imgs
(all_anchors, all_labels, all_label_weights, all_bbox_targets,
all_bbox_weights, pos_inds_list, neg_inds_list,
sampling_results_list) = multi_apply(
self._get_targets_single,
anchor_list,
valid_flag_list,
num_level_anchors_list,
batch_gt_instances,
batch_img_metas,
batch_gt_instances_ignore,
unmap_outputs=unmap_outputs)
# Get `avg_factor` of all images, which calculate in `SamplingResult`.
# When using sampling method, avg_factor is usually the sum of
# positive and negative priors. When using `PseudoSampler`,
# `avg_factor` is usually equal to the number of positive priors.
avg_factor = sum(
[results.avg_factor for results in sampling_results_list])
# split targets to a list w.r.t. multiple levels
anchors_list = images_to_levels(all_anchors, num_level_anchors)
labels_list = images_to_levels(all_labels, num_level_anchors)
label_weights_list = images_to_levels(all_label_weights,
num_level_anchors)
bbox_targets_list = images_to_levels(all_bbox_targets,
num_level_anchors)
bbox_weights_list = images_to_levels(all_bbox_weights,
num_level_anchors)
return (anchors_list, labels_list, label_weights_list,
bbox_targets_list, bbox_weights_list, avg_factor)
def _get_targets_single(self,
flat_anchors: Tensor,
valid_flags: Tensor,
num_level_anchors: List[int],
gt_instances: InstanceData,
img_meta: dict,
gt_instances_ignore: Optional[InstanceData] = None,
unmap_outputs: bool = True) -> tuple:
"""Compute regression, classification targets for anchors in a single
image.
Args:
flat_anchors (Tensor): Multi-level anchors of the image, which are
concatenated into a single tensor of shape (num_anchors ,4)
valid_flags (Tensor): Multi level valid flags of the image,
which are concatenated into a single tensor of
shape (num_anchors,).
num_level_anchors (List[int]): Number of anchors of each scale
level.
gt_instances (:obj:`InstanceData`): Ground truth of instance
annotations. It usually includes ``bboxes`` and ``labels``
attributes.
img_meta (dict): Meta information for current image.
gt_instances_ignore (:obj:`InstanceData`, optional): Instances
to be ignored during training. It includes ``bboxes`` attribute
data that is ignored during training and testing.
Defaults to None.
unmap_outputs (bool): Whether to map outputs back to the original
set of anchors.
Returns:
tuple: N is the number of total anchors in the image.
labels (Tensor): Labels of all anchors in the image with shape
(N,).
label_weights (Tensor): Label weights of all anchor in the
image with shape (N,).
bbox_targets (Tensor): BBox targets of all anchors in the
image with shape (N, 4).
bbox_weights (Tensor): BBox weights of all anchors in the
image with shape (N, 4)
pos_inds (Tensor): Indices of positive anchor with shape
(num_pos,).
neg_inds (Tensor): Indices of negative anchor with shape
(num_neg,).
sampling_result (:obj:`SamplingResult`): Sampling results.
"""
inside_flags = anchor_inside_flags(flat_anchors, valid_flags,
img_meta['img_shape'][:2],
self.train_cfg['allowed_border'])
if not inside_flags.any():
raise ValueError(
'There is no valid anchor inside the image boundary. Please '
'check the image size and anchor sizes, or set '
'``allowed_border`` to -1 to skip the condition.')
# assign gt and sample anchors
anchors = flat_anchors[inside_flags, :]
num_level_anchors_inside = self.get_num_level_anchors_inside(
num_level_anchors, inside_flags)
pred_instances = InstanceData(priors=anchors)
assign_result = self.assigner.assign(pred_instances,
num_level_anchors_inside,
gt_instances, gt_instances_ignore)
sampling_result = self.sampler.sample(assign_result, pred_instances,
gt_instances)
num_valid_anchors = anchors.shape[0]
bbox_targets = torch.zeros_like(anchors)
bbox_weights = torch.zeros_like(anchors)
labels = anchors.new_full((num_valid_anchors, ),
self.num_classes,
dtype=torch.long)
label_weights = anchors.new_zeros(num_valid_anchors, dtype=torch.float)
pos_inds = sampling_result.pos_inds
neg_inds = sampling_result.neg_inds
if len(pos_inds) > 0:
if self.reg_decoded_bbox:
pos_bbox_targets = sampling_result.pos_gt_bboxes
else:
pos_bbox_targets = self.bbox_coder.encode(
sampling_result.pos_priors, sampling_result.pos_gt_bboxes)
bbox_targets[pos_inds, :] = pos_bbox_targets
bbox_weights[pos_inds, :] = 1.0
labels[pos_inds] = sampling_result.pos_gt_labels
if self.train_cfg['pos_weight'] <= 0:
label_weights[pos_inds] = 1.0
else:
label_weights[pos_inds] = self.train_cfg['pos_weight']
if len(neg_inds) > 0:
label_weights[neg_inds] = 1.0
# map up to original set of anchors
if unmap_outputs:
num_total_anchors = flat_anchors.size(0)
anchors = unmap(anchors, num_total_anchors, inside_flags)
labels = unmap(
labels, num_total_anchors, inside_flags, fill=self.num_classes)
label_weights = unmap(label_weights, num_total_anchors,
inside_flags)
bbox_targets = unmap(bbox_targets, num_total_anchors, inside_flags)
bbox_weights = unmap(bbox_weights, num_total_anchors, inside_flags)
return (anchors, labels, label_weights, bbox_targets, bbox_weights,
pos_inds, neg_inds, sampling_result)
def get_num_level_anchors_inside(self, num_level_anchors, inside_flags):
"""Get the number of valid anchors in every level."""
split_inside_flags = torch.split(inside_flags, num_level_anchors)
num_level_anchors_inside = [
int(flags.sum()) for flags in split_inside_flags
]
return num_level_anchors_inside
|